Design of On-demand DBS in Movement Disorders using Machine Learning and the Need for Adaptive Learning

This thesis presents an automated tremor prediction algorithm based on modified Large memory storage and retrieval Neural Network (LNN-2), for the design of closed-loop Deep Brain Stimulation (DBS) system. The proposed method modifies the current open-loop paradigm of DBS to work on-demand by forecasting the onset of tremor in Parkinson’s Disease (PD) and Essential Tremor (ET) patients. Feedback provided by non-invasive physiological signals is used to drive the DBS in an on-off regime, stimulating the target region only when required. Such closed-loop DBS systems, thereby reduce the amount of stimulation applied to the brain and may also lead to improving the battery life, decreasing the risk of infection due to repetitive battery replacement surgeries. With the recent rise in wearable devices, we envisage a closed-loop system based on sEMG and acc signals for tremor-dominant PD and ET patients.